I've seen comments over the years that the engine nps reported by ChessGUI is understated, by quite a significant amount. In the years that I used it I never noticed that. Today I thought I'd investigate.
I took a pgn of a single opening position and started a tournament in Cutechess and ChessGUI one after the other. I set it for game in 2 hours so that I could snapshot the engine nps as white after a minute of thinking time. Here are the results:
Stockfish 17.1
836 nps - Cutechess
839 nps - ChessGUI
Caissa 1.22
2012 nps - Cutechess
2049 nps - ChessGUI
Integral V7
1255 nps - Cutechess
1264 nps - ChessGUI
As you can see, near enough identical. ChessGUI does not understate the engine nps.
The myth is just that - a myth - with no basis in reality at least for me. Your mileage may vary.
ChessGUI nps myth
Moderator: Ras
-
Modern Times
- Posts: 3702
- Joined: Thu Jun 07, 2012 11:02 pm
Re: ChessGUI nps myth
I don't know about Graham's live broadcasts. Is the web viewer showing a different nps to the GUI on his local machine ? He will know the answer to that, it may well be that that is where a variance occurs.
-
mar
- Posts: 2654
- Joined: Fri Nov 26, 2010 2:00 pm
- Location: Czech Republic
- Full name: Martin Sedlak
Re: ChessGUI nps myth
interesting
time and nodes come with WPV/BPV messages from the TLCS server (time in centiseconds and nodes)
might be interesting to compare the nps reported by the web viewer vs TLCV
what remains is to check the winboard.debug file generated by ChessGUI, but I very much doubt this is the case
time and nodes come with WPV/BPV messages from the TLCS server (time in centiseconds and nodes)
might be interesting to compare the nps reported by the web viewer vs TLCV
what remains is to check the winboard.debug file generated by ChessGUI, but I very much doubt this is the case